Input Selection And Structure Selection Of Soft Sensors Model Of Key Water Quality Variables In Wastewater Treatment:Set Membership Method

Nowadays,there is not only a shortage of water resources in China,but also more and more serious water pollution and environmental pollution.Because the wastewater treatment process involves a very complex reaction process,and there are many microorganisms involved in the whole reaction process,so it is difficult for us to measure the effluent quality indicator scientifically and effectively.However,it is precisely because some water quality indicators play an important role in the operation of the wastewater treatment plant.Therefore,in the process of wastewater treatment,it is particularly important to monitor the water quality content in real time and accurately,especially to predict an important indicator in real time.In the real-time prediction of a variable,the selection of secondary variables is also essential.Therefore,in this paper,the idea of set membership identification is used to select input variables and adjust the hidden layer structure of neural network.The selection of input variables is divided into two parts.First,the initial secondary variables are selected by stepwise regression,then the initial secondary variables are further screened by the idea of set membership,and finally the appropriate secondary variables are obtained.According to the input and output data,the appropriate network structure is selected and the interval prediction model is established.However,because the network structure may have a certain degree of redundancy,this paper also uses the method of set membership structure reduction to adjust the hidden layer structure of the neural network,and optimizes the structure of the neural network under the condition of ensuring the same prediction accuracy.The research direction and main contents of this paper are as follows:(1)Input selection of soft sensors model for key water quality indicators of wastewater treatment:At present,there are many methods to select the secondary variables of key water quality indicators of wastewater treatment,including partial least squares(PLS)and principal component analysis(PCA).However,this method often loses the original characteristics of the data,recombines the input variables to get new variables,so that the new variables contain the information of one or more original data,and changes the original characteristics of the data;which retains the original characteristics of the data.In this paper,the method of stepwise regression is used to select the secondary variable.First,the regression coefficient can be estimated by using the least square estimation,and a linear regression model is established.Given a significance level α,one of the variables can be deleted each time until it meets the conditions.At the same time,the variables obtained by stepwise regression are further screened by set membership variable selection,that is,one or more variables are selected from the existing secondary variables for verification each time,so that the error obtained by each identification is as small as possible,and finally the group of variables with the smallest error is selected as the input variable of the model,and then a new soft measurement model is constructed by using the selected input and output variables.(2)Soft sensors model structure selection of key water quality indicators of wastewater treatment:After selecting appropriate secondary variables,the center and width of the neural network of RBF is calculated by K-means mean clustering algorithm,combined with the bounded assumption of modeling error,the interval estimation models under different secondary variables are established to obtain the reliability interval estimation of the variables to be predicted.At the same time,in order to simplify the structure of the network model,the hidden layer of the constructed network is adjusted by using the method of set membership structure deletion,the redundant nodes of the hidden layer are deleted,and the structure of the network model is optimized on the basis of reliability interval estimation.In the experiment,the input of 2-dimension is selected to predict the effluent BOD,and a simplified neural network model is established.The experimental results show the effectiveness and feasibility of the soft sensor model.(3)Design of interval prediction system for key water quality parameters of sewage treatment:This paper designs and develops a soft sensors system based on the water quality index of wastewater treatment,which includes start-up module,user login module and main interface operation module.In the system design process,we use JAVA and MATLAB mixed programming technology to realize the MATLAB program calling RBF neural network in the interface,so that the key parameters of the water can be displayed and saved by the training and prediction results of the neural network.This design mainly shows the visual effect chart of neural network variable selection and structure deletion,and realizes the interval prediction and estimation of effluent quality indicator BOD.At the same time,the whole process of soft sensors model is visualized,and it is very convenient for users to understand the running process of wastewater treatment in real time.